This paper proposes a machine learning based method to predict the ‘queue-time’ of a target lot with time-constraints among processing steps. Since time-constraints among processing steps exist for quality assurance, it is very important to meet the time-constraints in Fab operations. To do so, we need to have an accurate prediction model for the ‘queue-time’ of a target lot. This paper identifies two categories of predictor variables; 1) work-in-process related variables, and 2) dispatching rule related variables. Since the quality of the prediction model depends on the quality of predictor variables, we need to carefully select predictor variables by considering their explanatory powers and multi-collinearities among them. In this paper, we employ the genetic algorithm for the section of best predictor variables. The prediction model is a fully-connected deep learning model, and the demonstration shows the performance of the model.
